Seminar by Joe MacInnes 29/03/2024

vendredi 29 mars 2024 à 14:30
Publié le 26/03/2024

Friday 29th March 2024, 14h30 – Salle Henri Gastaut

Joe MacInnes (Swansea University, UK) invited by Liuba Ardasheva and Laurent Perrinet

Casting a wide (neural) net: models and simulations of eye movements and attention

Abstract: Eye movements are an excellent proxy for visual attention, and offer a rich source of behavioural data. Decades of neuroscience and experimental results have provided many interesting artefacts that hint at underlying attentional mechanisms. Models and simulations of attention allow us the opportunity to implement our best theories and test them against a wide variety of experimental and imaging results. Simulations of human eye movements, for example, can predict where we allocate attention, the temporal distributions of saccades and even the presence of attentional artifacts like Inhibition of Return, errors, anticipations and even virtual TMS lesions. This talk will cover a couple of recent models that simulate attention and eye movements using deep learning neural nets, spiking network layers and diffusion models to simulate aspects, quirks and mechanisms of human attention.